This article about recursive self-improvement from Anthropic is fascinating. Lots of good info in it. But I’m not sure I can take this part seriously:
We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.
These are the folks who were worried back in 2019 about GPT-2, a model that in hindsight seems quaint, not dangerous. Maybe they’re right this time. At the risk of being overly cynical, it’s also pretty convenient if your competition pauses development while you’ve got billions of dollars coming in, revenue that would continue during a pause.
For better or worse, we are barreling forward. Government needs to respond, on energy, regulation, and basic income. I’m not sure a pause would even help much. The reason society is so far behind is a lack of imagination for what the future could look like, not a lack of time.
Some people who view AI as a bubble think that all the companies are going to vanish when the bubble pops. That’s not going to happen. Many companies will fail, but the few biggest players are likely here to stay, and they will have enormous power.
The best course of action is to channel some of those billions of dollars in funding and revenue into infrastructure and new social programs. A pause is not only unrealistic, it might get in the way of a once-a-generation chance to build something great. Our biggest problem isn’t speed, it’s vision.
